DOCSLIB.ORG

Deep Learning for User Comment Moderation

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Deep Learning for User Comment Moderation Deep Learning for User Comment Moderation John Pavlopoulos Prodromos Malakasiotis Ion Androutsopoulos StrainTek StrainTek Department of Informatics Athens, Greece Athens, Greece Athens University of Economics [email protected] [email protected] and Business, Greece [email protected] Abstract x x Experimenting with a new dataset of 1.6M ✓ ✓ user comments from a Greek news portal ? ? and existing datasets of English Wikipedia ? ? x x x comments, we show that an RNN outper- x x forms the previous state of the art in mod- ✓ ✓ ✓ eration. A deep, classification-specific at- ✓ ✓ ✓ tention mechanism improves further the overall performance of the RNN. We also Figure 1: Semi-automatic moderation. compare against a CNN and a word-list baseline, considering both fully automatic and semi-automatic moderation. comments from a Greek sports portal (Gazzetta), which we make publicly available.2 Furthermore, 1 Introduction we provide word embeddings pre-trained on 5.2M User comments play a central role in social me- comments from the same portal. We also exper- dia and online discussion fora. News portals and iment on the datasets of Wulczyn et al. (2017), blogs often also allow their readers to comment which contain English Wikipedia comments la- in order to get feedback, engage their readers, beled for personal attacks, aggression, toxicity. and build customer loyalty. User comments, how- In a fully automatic scenario, a system directly ever, and more generally user content can also accepts or rejects comments. Although this sce- be abusive (e.g., bullying, profanity, hate speech). nario may be the only available one, e.g., when Social media are increasingly under pressure to portals cannot afford moderators, it is unrealistic combat abusive content. News portals also suf- to expect that fully automatic moderation will be fer from abusive user comments, which damage perfect, because abusive comments may involve their reputation and make them liable to fines, e.g., irony, sarcasm, harassment without profanity etc., when hosting comments encouraging illegal ac- which are particularly difficult for machines to tions. They often employ moderators, who are fre- handle. When moderators are available, it is more quently overwhelmed by the volume of comments. realistic to develop semi-automatic systems to as- Readers are disappointed when non-abusive com- sist rather than replace them, a scenario that has ments do not appear quickly online because of not been considered in previous work. Comments moderation delays. Smaller news portals may be for which the system is uncertain (Fig.1) are unable to employ moderators, and some are forced shown to a moderator to decide; all other com- to shut down their comments sections entirely.1 ments are accepted or rejected by the system. We We examine how deep learning (Goodfellow discuss how moderation systems can be tuned, de- et al., 2016; Goldberg, 2016) can be used to mod- pending on the availability and workload of mod- erate user comments. We experiment with a new erators. We also introduce additional evaluation dataset of approx. 1.6M manually moderated user 2The portal is http://www.gazzetta.gr/. In- 1See, for example, http://niemanreports.org/ structions to obtain the Gazzetta data will be posted at http: articles/the-future-of-comments/. //nlp.cs.aueb.gr/software.html. 25 Proceedings of the First Workshop on Abusive Language Online, pages 25–35, Vancouver, Canada, July 30 - August 4, 2017. c 2017 Association for Computational Linguistics Dataset/Split Accepted Rejected Total G-TRAIN-L, called G-TRAIN-S. An additional set G-TRAIN-L 960,378 (66%) 489,222 (34%) 1,45M G-TRAIN-S 67,828 (68%) 32,172 (32%) 100,000 of 60,900 comments (Oct. 7 to Nov. 11, 2016) G-DEV 20,236 (68%) 9,464 (32%) 29,700 was split to development (G-DEV, 29,700 com- G-TEST-L 20,064 (68%) 9,636 (32%) 29,700 ments), large test (G-TEST-L, 29,700), and small G-TEST-S 1,068 (71%) 432 (29%) 1,500 G-TEST-S-R 1,174 (78%) 326 (22%) 1,500 test set (G-TEST-S, 1,500). Gazzetta’s moderators W-ATT-TRAIN 61,447 (88%) 8,079 (12%) 69,526 (2 full-time, plus journalists occasionally helping) W-ATT-DEV 20,405 (88%) 2,755 (12%) 23,160 are occasionally instructed to be stricter (e.g., dur- W-ATT-TEST 20,422 (88%) 2,756 (12%) 23,178 W-TOX-TRAIN 86,447 (90%) 9,245 (10%) 95,692 ing violent events). To get a more accurate view W-TOX-DEV 29,059 (90%) 3,069 (10%) 32,128 of performance in normal situations, we manually W-TOX-TEST 28,818 (90%) 3,048 (10%) 31,866 re-moderated (labeled as ‘accept’ or ‘reject’) the Table 1: Statistics of the datasets used. comments of G-TEST-S, producing G-TEST-S-R. The reject ratio is approximately 30% in all sub- sets, except for G-TEST-S-R where it drops to 22%, measures for the semi-automatic scenario. because there are no occasions where the modera- On both Gazzetta and Wikipedia comments and tors were instructed to be stricter in G-TEST-S-R. for both scenarios (automatic, semi-automatic), Each G-TEST-S-R comment was re-moderated we show that a recursive neural network (RNN) by 5 annotators. Krippendorff’s (2004) alpha was outperforms the system of Wulczyn et al. (2017), 0.4762, close to the value (0.45) reported by Wul- the previous state of the art for comment modera- czyn et al. (2017) for Wikipedia comments. Using tion, which employed logistic regression (LR) or a Cohen’s Kappa (Cohen, 1960), the mean pairwise multi-layered Perceptron (MLP). We also propose agreement was 0.4749. The mean pairwise per- an attention mechanism that improves the over- centage of agreement (% of comments each pair all performance of the RNN. Our attention differs of annotators agreed on) was 81.33%. Cohen’s from most previous ones (Bahdanau et al., 2015; Kappa and Krippendorff’s alpha lead to moder- Luong et al., 2015) in that it is used in text clas- ate scores, because they account for agreement by sification, where there is no previously generated chance, which is high when there is class imbal- output subsequence to drive the attention, unlike ance (22% reject, 78% accept in G-TEST-S-R). sequence-to-sequence models (Sutskever et al., We also provide 300-dimensional word em- 2014). In effect, our attention mechanism detects beddings, pre-trained on approx. 5.2M comments the words of a comment that affect mostly the clas- (268M tokens) from Gazzetta using WORD2VEC sification decision (accept, reject), by examining (Mikolov et al., 2013a,b).3 This larger dataset can- them in the context of the particular comment. not be used to train classifiers, because most of its Our main contributions are: (i) We release a comments are from a period (before 2015) when new dataset of 1.6M moderated user comments. Gazzetta did not employ moderators. (ii) We are among the first to apply deep learning to user comment moderation, and we show that an 2.2 Wikipedia datasets RNN with a novel classification-specific attention Wulczyn et al. (2017) created three datasets con- mechanism outperforms the previous state of the taining English Wikipedia talk page comments. art. (iii) Unlike previous work, we also consider Attacks dataset: This dataset contains approx. a semi-automatic scenario, along with threshold 115K comments, which were labeled as personal tuning and evaluation measures for it. attacks (reject) or not (accept) using crowdsourc- ing. Each comment was labeled by at least 10 an- 2 Datasets notators. Inter-annotator agreement, measured on We first discuss the datasets we used, to help ac- a random sample of 1K comments using Krippen- quaint the reader with the problem. dorff’s (2004) alpha, was 0.45. The gold label of each comment is determined by the majority of an- 2.1 Gazzetta dataset notators, leading to binary labels (accept, reject). There are approx. 1.45M training comments (cov- Alternatively, the gold label is the percentage of ering Jan. 1, 2015 to Oct. 6, 2016) in the Gazzetta annotators that labeled the comment as ‘accept’ dataset; we call them G-TRAIN-L (Table1). Some 3We used CBOW, window size 5, min. term freq. 5, nega- experiments use only the first 100K comments of tive sampling, obtaining a vocabulary size of approx. 478K. 26 (or ‘reject’), leading to probabilistic labels.4 The bag of word n-grams (n 2, each comment be- ≤ dataset is split in three parts (Table1): training ( W- comes a bag containing its 1-grams and 2-grams) ATT-TRAIN, 69,526 comments), development (W- or a bag of character n-grams (n 5, each com- ≤ ATT-DEV, 23,160), and test (W-ATT-TEST, 23,178 ment becomes a bag containing character 1-grams, comments). In all three parts, the rejected com- . , 5-grams). DETOX can rely on a logistic regres- ments are 12%, but this ratio is artificial (in effect, sion (LR) or multi-layer Perceptron (MLP) clas- Wulczyn et al. oversampled comments posted by sifier, and use binary or probabilistic gold labels banned users), unlike Gazzetta subsets where the (Section 2.2) during training. We used the DETOX truly observed accept/reject ratios are used. implementation of Wulczyn et al. and the same Toxicity dataset: This dataset was created like grid search to tune the hyper-parameters that se- the previous one, but contains more comments lect word or character n-grams, classifier (LR or (159,686), now labeled as toxic (reject) or not (ac- MLP), and gold labels (binary or probabilistic). cept). Inter-annotator agreement was not reported.
Load more

Recommended publications
  • ``At the End of the Day Facebook Does What It Wants'': How Users
    “At the End of the Day Facebook Does What It Wants”: How Users Experience Contesting Algorithmic Content Moderation KRISTEN VACCARO, University of Illinois Urbana-Champaign CHRISTIAN SANDVIG, University of Michigan KARRIE KARAHALIOS, University of Illinois Urbana-Champaign Interest has grown in designing algorithmic decision making systems for contestability. In this work, we study how users experience contesting unfavorable social media content moderation decisions. A large-scale online experiment tests whether different forms of appeals can improve users’ experiences of automated decision making. We study the impact on users’ perceptions of the Fairness, Accountability, and Trustworthiness of algorithmic decisions, as well as their feelings of Control (FACT). Surprisingly, we find that none of the appeal designs improve FACT perceptions compared to a no appeal baseline. We qualitatively analyze how users write appeals, and find that they contest the decision itself, but also more fundamental issues like thegoalof moderating content, the idea of automation, and the inconsistency of the system as a whole. We conclude with suggestions for – as well as a discussion of the challenges of – designing for contestability. CCS Concepts: • Human-centered computing → Human computer interaction (HCI); Social media. Additional Key Words and Phrases: content moderation; algorithmic experience ACM Reference Format: Kristen Vaccaro, Christian Sandvig, and Karrie Karahalios. 2020. “At the End of the Day Facebook Does What It Wants”: How Users Experience Contesting Algorithmic Content Moderation. Proc. ACM Hum.-Comput. Interact. 4, CSCW2, Article 167 (October 2020), 22 pages. https://doi.org/10.1145/3415238 1 INTRODUCTION As algorithmic decision making systems become both more prevalent and more visible, interest has grown in how to design them to be trustworthy, understandable and fair.
    [Show full text]
  • Download Paper
    Lawless: the secret rules that govern our digital lives (and why we need new digital constitutions that protect our rights) Submitted version. Forthcoming 2019 Cambridge University Press. Nicolas P. Suzor Table of Contents Part I: a lawless internet Chapter 1. The hidden rules of the internet ............................................................................................. 6 Process matters ....................................................................................................................................................... 12 Chapter 2. Who makes the rules?.......................................................................................................... 17 Whose values apply? ............................................................................................................................................... 22 The moderation process .......................................................................................................................................... 25 Bias and accountability ........................................................................................................................................... 28 Chapter 3. The internet’s abuse problem ............................................................................................... 41 Abuse reflects and reinforces systemic inequalities ................................................................................................ 50 Dealing with abuse needs the involvement of platforms .......................................................................................
    [Show full text]
  • Content Moderation
    Robert Lewington (Twitch) & The Fair Play Alliance Executive Steering Committee April 2021 Being ‘Targeted’ about Content Moderation: Strategies for consistent, scalable and effective response to Disruption & Harm 1 Content Moderation: Best Practices for Targeted Reporting & reactive UGC Management At Scale March 2021 Abstract This document provides replicable best practice information on how to moderate User-Generated Content (UGC) in social applications or services (including digital media and video games). Its focus is on reactive moderation, a central component of the growing content moderation toolkit where a service provider responds to reports submitted by users of its service regarding UGC that may violate its Terms of Service. Specifically, the document explores and advocates for a ‘targeted’ approach to the creation of reporting mechanisms. This allows users to closely identify the specific infraction, utilise evidence of the infraction—access to which is facilitated as part of the design of the reporting process—enabling consistent content moderation at scale. Note, however, that we will also make passing-reference to pre, post and proactive (see Appendix A) moderation approaches. Specifics of how best to tailor these best practices to a particular application or service will differ based on various parameters, including: type of service (social media, video game etc.); type of media (text, image, audio, video etc.); sharing mechanism (feed/gallery, avatar, communication etc.); persistence (ephemeral vs. static/umutable) and others, and therefore this document should be considered a set of high-level instructive principles rather than prescriptive guidelines. Contents i. Background ii. The Content Moderation Flywheel ○ Community Guidelines/Code of Conduct ○ Targeted reporting ■ Context ○ Scalable Content Moderation ○ Education ○ Technology iii.
    [Show full text]
  • Deeper Attention to Abusive User Content Moderation
    Deeper Attention to Abusive User Content Moderation John Pavlopoulos Prodromos Malakasiotis Ion Androutsopoulos Straintek, Athens, Greece Straintek, Athens, Greece Athens University of Economics [email protected] [email protected] and Business, Greece [email protected] Abstract suffer from abusive user comments, which dam- age their reputations and make them liable to fines, Experimenting with a new dataset of 1.6M e.g., when hosting comments encouraging illegal user comments from a news portal and an actions. They often employ moderators, who are existing dataset of 115K Wikipedia talk frequently overwhelmed, however, by the volume page comments, we show that an RNN op- and abusiveness of comments.3 Readers are dis- erating on word embeddings outpeforms appointed when non-abusive comments do not ap- the previous state of the art in moderation, pear quickly online because of moderation delays. which used logistic regression or an MLP Smaller news portals may be unable to employ classifier with character or word n-grams. moderators, and some are forced to shut down We also compare against a CNN operat- their comments sections entirely. ing on word embeddings, and a word-list We examine how deep learning (Goodfellow baseline. A novel, deep, classification- et al., 2016; Goldberg, 2016, 2017) can be em- specific attention mechanism improves the ployed to moderate user comments. We experi- performance of the RNN further, and can ment with a new dataset of approx. 1.6M manually also highlight suspicious words for free, moderated (accepted or rejected) user comments without including highlighted words in the from a Greek sports news portal (called Gazzetta), training data.
    [Show full text]
  • Recommendations for the Facebook Content Review Board
    STANFORD Policy Practicum: Creating a Social Media Oversight Board Recommendations for the Facebook Content Review Board 2018-19 PRACTICUM RESEARCH TEAM: Shaimaa BAKR, Ph.D. Electrical Engineering ‘20 Madeline MAGNUSON, J.D. ’20 Fernando BERDION-DEL VALLE, J.D. ’20 Shawn MUSGRAVE, J.D. ’21 Isabella GARCIA-CAMARGO, B.S.’20 Ashwin RAMASWAMI, B.S. ‘21 Julia GREENBERG, J.D. ’19 Nora TAN, B.S. ’19 Tara IYER, B.S.’19 Marlena WISNIAK, LL.M. ’19 Alejandra LYNBERG, J.D. ’19 Monica ZWOLINSKI, J.D. ’19 INSTRUCTORS: Paul BREST, Faculty Director, Law and Policy Lab Daniel HO, William Benjamin Scott and Luna M. Scott Professor of Law Nathaniel PERSILY, James B. McClatchy Professor of Law Rob REICH, Faculty Director, Center for Ethics in Society TEACHING ASSISTANT: Liza STARR, J.D./M.B.A. ’21 SPRING POLICY CLIENT: Project on Democracy and the Internet, Stanford Center on Philanthropy and Civil Society 2019 559 Nathan Abbot Way Stanford, CA https://law.stanford.edu/education/only-at-sls/law-policy-lab// Contents Executive Summary ....................................................................................3 Introduction ................................................................................................8 I. The Need for Governing Principles ..........................................................9 II. Structure and Membership ...................................................................10 A. Selection Process for Board Members ..............................................................10 B. Membership Criteria .............................................................................................12
    [Show full text]
  • Online Communities As Semicommons
    The Virtues of Moderation: Online Communities as Semicommons James Grimmelmann* IP Scholars workshop draft: not intended for broad distribution or citation I. INTRODUCTION................................................................................................................................................2 II. BACKGROUND ..................................................................................................................................................4 A. PUBLIC AND PRIVATE GOODS ...........................................................................................................................4 B. THE TRAGIC STORY ..........................................................................................................................................6 C. THE COMEDIC STORY .......................................................................................................................................9 D. LAYERING.......................................................................................................................................................12 III. ONLINE SEMICOMMONS .............................................................................................................................14 A. A FORMAL MODEL .........................................................................................................................................14 B. COMMONS VIRTUES........................................................................................................................................17
    [Show full text]
  • Online Comment Moderation: Emerging Best Practices a Guide to Promoting Robust and Civil Online Conversation
    Online comment moderation: emerging best practices A guide to promoting robust and civil online conversation www.wan-ifra.org Online comment moderation: emerging best practices by Emma Goodman RESEARCHERS: SUPPORTED BY: Federica Cherubini, Alexandra Waldhorn Media Program of the Open Society Foundations EDITORS: www.opensocietyfoundations.org Alexandra Waldhorn, Amy Had!eld PUBLISHED BY: WAN-IFRA Washingtonplatz 1 64287 Darmstadt, Germany Tel. +49 6151 7336 / Fax +49 6151 733800 www.wan-ifra.org CEO: Vincent Peyrègne DIRECTOR OF PUBLICATIONS: Dean Roper Published by: the World Association of Newspapers (WAN-IFRA) WORLD EDITORS FORUM DIRECTOR: Cherilyn Ireton © 2013 World Association of Newspapers and News Publishers Table of contents Introduction ............................................ 5 The status of comments .........................17 The moderation process ........................ 21 Issues and challenges .............................51 Best practices ........................................ 61 What’s next? ......................................... 65 Conclusion ............................................ 71 Online comment moderation: challenges and opportunities Image © sxc.hu | Miguel Ugalde 4 A guide to promoting robust and civil online conversation Introduction In many parts of the globe, online com- consequence-free behaviour and a chance ments have become an essential ingredient to defy social norms, or maybe it’s a factor of a thriving news publication: readers feel of the structure of online conversations, that they have
    [Show full text]
  • Facebook's "Oversight Board:" Move Fast with Stable Infrastructure and Humility
    NORTH CAROLINA JOURNAL OF LAW & TECHNOLOGY Volume 21 Issue 1 Article 2 10-1-2019 Facebook's "Oversight Board:" Move Fast with Stable Infrastructure and Humility Evelyn Douek Follow this and additional works at: https://scholarship.law.unc.edu/ncjolt Part of the Law Commons Recommended Citation Evelyn Douek, Facebook's "Oversight Board:" Move Fast with Stable Infrastructure and Humility, 21 N.C. J.L. & TECH. 1 (2019). Available at: https://scholarship.law.unc.edu/ncjolt/vol21/iss1/2 This Article is brought to you for free and open access by Carolina Law Scholarship Repository. It has been accepted for inclusion in North Carolina Journal of Law & Technology by an authorized editor of Carolina Law Scholarship Repository. For more information, please contact [email protected]. NORTH CAROLINA JOURNAL OF LAW & TECHNOLOGY VOLUME 21, ISSUE 1: OCTOBER 2019 FACEBOOK’S “OVERSIGHT BOARD:” MOVE FAST WITH STABLE INFRASTRUCTURE AND HUMILITY Evelyn Douek* Facebook’s proposed Oversight Board is one of the most ambitious constitution-making projects of the modern era. With pre- existing governance of tech platforms delegitimized in the ongoing “techlash,” this represents a pivotal moment when new constitutional forms can emerge that will shape the future of online discourse. For all the potential of the Facebook Oversight Board (FOB), there are many things it cannot be. It will not hear a sufficient proportion of Facebook’s content moderation cases to be a meaningful response to calls for greater due process in individual decisions. Nor will it be able to become a font of globally accepted speech norms for the worldwide platform.
    [Show full text]
  • Content Moderation
    EVERYTHING IN MODERATI N PLATFORMS, COMMUNITIES AND USERS IN A HEALTHY ONLINE ENVIRONMENT ALEX KRASODOMSKI-JONES OCTOBER 2020 Open Access. Some rights reserved. Open Access. Some rights reserved. As the publisher of this work, Demos wants to encourage the circulation of our work as widely as possible while retaining the copyright. We therefore have an open access policy which enables anyone to access our content online without charge. Anyone can download, save, perform or distribute this work in any format, including translation, without written permission. This is subject to the terms of the Creative Commons By Share Alike licence. The main conditions are: • Demos and the author(s) are credited including our web address www.demos.co.uk • If you use our work, you share the results under a similar licence A full copy of the licence can be found at https://creativecommons.org/licenses/ by-sa/3.0/legalcode You are welcome to ask for permission to use this work for purposes other than those covered by the licence. Demos gratefully acknowledges the work of Creative Commons in inspiring our approach to copyright. To find out more go to www.creativecommons.org This project was supported by GCHQ Published by Demos October 2020 © Demos. Some rights reserved. 15 Whitehall, London, SW1A 2DD T: 020 3878 3955 [email protected] www.demos.co.uk CONTENTS ACKNOWLEDGEMENTS PAGE 4 INTRODUCTION PAGE 5 ONLINE HARMS PAGE 8 LEGAL vs ILLEGAL PART 1 PAGE 9 PLATFORM LEVEL MODERATION AND THE TERMS OF SERVICE PART 2 PAGE 14 COMMUNITY MODERATION PART 3 PAGE 22 SELF-MODERATION AND ONLINE CULTURES AND NORMS PART 4 PAGE 25 RECOMMENDATIONS RECOMMENDED READING PAGE 26 3 ACKNOWLEDGEMENTS The generosity shown by the contributors to this paper in giving up their time to explore how constitutional law, criminology, social control and policing may shed light on the challenges of the online world was remarkable.
    [Show full text]
  • Controlling the Conversation: the Ethics of Social Platforms and Content Moderation
    Controlling the Conversation: The Ethics of Social Platforms and Content Moderation April 2018 Priyanjana Bengani With introduction by Mike Ananny and conclusion by Emily Bell Platforms and Publishers: Policy Exchange Forum III February 23, 2018 | University of Southern California Annenberg Hosted by Tow Center for Digital Journalism at Columbia University and the Annenberg Innovation Lab Controlling the Conversation 1 Executive Summary 3 Introduction 5 Discussion I: Ethics of Moderation 8 Ethics: A Short Introduction, or Democracy Without Dialogue 8 Journalism Ethics versus Content Moderation Ethics 9 The Role of Publishers versus The Role of Platforms 10 Key Challenges and Opportunities 11 Discussion II: Moderation Tools 12 Objectives of Stakeholders: Newsrooms, Platforms, Advertisers, Users 13 Automated Moderation versus Human Moderation 14 Key Challenges and Opportunities 16 Discussion III: Technological Challenges 17 Designing and Implementing a Technology-Driven Solution 17 Platforms, Editorial Oversight, and the Laws 19 Key Challenges and Opportunities 20 Conclusion 21 The Policy Exchange Forums are a critical component of the Tow Center’s Platforms and Publishers research project. In these sessions, participants representing both the platforms and publishing sides of the news industry can engage on issues related to the ethical and civic values of journalism. The forum focuses on the relationships between technology, business, journalism, and ethics, and brings together diverse stakeholders to discuss current issues and surface potential new ones. The project is underwritten by the John D. and Catherine T. MacArthur Foundation, with additional support by the John S. and James L. Knight Foundation, the Foundation to Promote Open Society, and The Abrams Foundation. Controlling the Conversation 2 Executive Summary With social platforms’ prevailing dominance, there are numerous debates around who owns information, content, and the audience itself: the publisher, or the platform where the content is discovered—or not discovered, as the case may be.
    [Show full text]
  • Reimagining Content Moderation and Safeguarding Fundamental Rights a Study on Community-Led Platforms
    REIMAGINING CONTENT MODERATION AND SAFEGUARDING FUNDAMENTAL RIGHTS A STUDY ON COMMUNITY-LED PLATFORMS Ben Wagner, Johanne Kübler, Eliška Pírková, Rita Gsenger, Carolina Ferro May 2021 Acknowledgements This study was commissioned by the Greens/EFA group in the European Parliament. The authors wish to thank the European Greens/EFA team, especially Kirsten Fiedler, Narmine Abou Bakari, and Eszter Bako, for their fantastic support. We would also like to acknowledge the valuable contribution of the people we interviewed during the process of elaboration of this study, without whom this study would not have been possible. Thank you for dedicating your time to reply to our questions, even in such uncertain and difficult times. Table of contents 2 ACKNOWLEDGEMENTS 4 INTRODUCTION 6 CONTENT GOVERNANCE THEORETICAL FRAMEWORK AND METHODOLOGICAL APPROACH 7 I. Theoretical framework 11 II. Research design, methodology and case selection 14 OVERVIEW OF THE CASES 14 I. Wikipedia 16 II. diaspora* 18 III. Mastodon 19 IV. Der Standard 20 V. Slashdot 22 VI. Advantages and disadvantages of different approaches 24 CONCRETE ALTERNATIVES BASED ON EMPIRICAL DATA 24 I. Accountability approach to increase rule compliance 25 II. Durable pseudonyms 27 III. Ranking and rating 28 IV. Forum design changes 29 POLICY RECOMMENDATIONS 33 FINAL CONSIDERATIONS 35 REFERENCES 44 ABBREVIATIONS 45 ANNEXES 45 I. Semi-structured interview questionnaire 45 II. Lists of interviews 46 III. Summary of interviews 3 Introduction As millions of people log into social media platforms like Facebook, Twitter, and Reddit every day, the rules by which these platforms are governed increasingly determine how we interact with each other, and they shape the possibilities and nature of public discourse (Suzor 2020).
    [Show full text]
  • Arxiv:2101.07183V2 [Cs.SI] 13 Apr 2021 Tially Harmful Information
    “I Won the Election!”: An Empirical Analysis of Soft Moderation Interventions on Twitter* Savvas Zannettou Max Planck Institute for Informatics [email protected] Abstract Over the past few years, there is a heated debate and serious public concerns regarding online content moderation, censor- ship, and the principle of free speech on the Web. To ease these concerns, social media platforms like Twitter and Facebook refined their content moderation systems to support soft mod- eration interventions. Soft moderation interventions refer to warning labels attached to potentially questionable or harmful content to inform other users about the content and its nature Figure 1: An example of a soft moderation intervention on Twitter. while the content remains accessible, hence alleviating con- cerns related to censorship and free speech. In this work, we perform one of the first empirical studies on soft moderation interventions on Twitter. Using a mixed- be performed in a timely manner to ensure that harmful con- methods approach, we study the users who share tweets with tent is removed fast and only a small number of users are ex- warning labels on Twitter and their political leaning, the en- posed to harmful content. This is a tough challenge given the gagement that these tweets receive, and how users interact scale of modern social media platforms like Twitter and Face- with tweets that have warning labels. Among other things, we book. Second, content moderation should be consistent and find that 72% of the tweets with warning labels are shared by fair across users. Finally, content moderation should ensure Republicans, while only 11% are shared by Democrats.
    [Show full text]